Model cardKnowledgator
GLiClass large v3.0 model card
Measured on: 14-dataset zero-shot classification average (F1)
- Zero-shot F1 average
- 0.676
Author-reported; standard text-classification sets, not decisions.
GLiClass base v3.0
KnowledgatorOpen weights
Knowledgator's size/accuracy pick; the best-calibrated GLiClass row once the payload is right.
- typed-decisions
- —
- Metatext suite
- 0.606
- ECE untuned
- 0.146
- p50 per case
- 300 ms
not run
5/7 over baseline
four CPU cores
Specifications
- Checkpoint
- knowledgator/gliclass-base-v3.0
- Publisher
- Knowledgator
- Mechanism
- Label-relevance classifier
- Backbone
- DeBERTa-v3-base
- Parameters
- 187M
- License
- Apache 2.0
- Runs as
- open weights, self-hosted
- Library
- gliclass
- Primitives
- Choice, Noul
- Pricing
- none — your compute
Metatext decisionsrank 7 of 21
Full leaderboard →| Configuration | Mode | Decisions | Beats baseline | ECE |
|---|---|---|---|---|
| criteria + single-hypothesis Noul (correct payload)headline | serial | 0.606 | 5/7 | 0.146 |
| criteria folded into labels | serial | 0.388 | 3/7 | 0.454 |
| default payload | serial | 0.381 | 2/7 | 0.518 |
Per-question accuracy — criteria + single-hypothesis Noul (correct payload), serial mode, 120 judgments
| Task | Question | Type | Accuracy | Baseline | Beats? |
|---|---|---|---|---|---|
| agent_traces | outcome | choice | 0.429 | 0.357 | ✅ |
| agent_traces | needs_human | noul | 0.786 | 0.714 | ✅ |
| customer_service | department | choice | 0.700 | 0.400 | ✅ |
| customer_service | refund_requested | noul | 0.950 | 0.700 | ✅ |
| customer_service | frustration | score | 0.200 | 0.500 | ❌ |
| security_incidents | severity | choice | 0.500 | 0.312 | ✅ |
| security_incidents | actively_exploited | noul | 0.688 | 0.688 | ❌ |
Published numbers
Reported by the authors or by third parties, on their own suites. Not comparable to the leaderboard above; attributed so you can check them.
Model cardFastino · 2025-07
GLiNER2 paper (arXiv 2507.18546)
Measured on: Seven zero-shot classification sets
- Reported comparator (GLiClass base v1.0)
- 0.63; DeBERTa-v3 zero-shot 0.69; GPT-4o 0.84
Peer-visible paper; the 205M model.
Frequently asked questions
How does GLiClass base v3.0 compare to Jev?
On the Metatext suite GLiClass base v3.0 scores 0.606 against Jev's 0.863.
Is GLiClass base v3.0 open source?
Yes. knowledgator/gliclass-base-v3.0 is released under Apache 2.0 and runs on your own hardware with the gliclass library.
Can GLiClass base v3.0 extract entities?
No. GLiClass base v3.0 answers questions about text and never returns spans; use a GLiNER-family model for extraction.